“People in a hurry will imitate more readily than people at leisure. Hustling thus tends to produce uniformity.” — Eric Hoffer
Last month I wrote about waking up at 4 a.m. trying to get to the bottom of the to-do list before baby #2 arrives. Well, I got to the bottom of the list and will probably be at the hospital when this goes out. Wish us luck!
Hope your summer/winter is going well - On to the links.
The Nature of Order, Book One: The Phenomenon of Life
by Christopher Alexander
Christopher Alexander, the architect best known for A Pattern Language, spent the last decades of his life on a four-volume work called The Nature of Order. It is very good. Book One, The Phenomenon of Life, lays the foundation: some buildings, objects, and places feel more alive than others, that feeling is surprisingly consistent from person to person, and it comes from recurring structural properties. The core unit is what he calls a center, a region that draws strength from everything around it and gives strength back.
A center is less a thing than a field. A fountain draws its life from the plaza around it, the plaza from the buildings that frame it, the buildings from their own windows and doors. They prop each other up and rise to life at the same time. That is why a place can feel alive without any single part being remarkable — the life is in the way the parts hold one another up, not in the parts themselves.
My notes on this book are bordering on embarrassingly enthusiastic. The one I keep coming back to: we are surrounded by complex systems, but we talk about them in a complicated way, through a mechanical metaphor. That is my core frustration with business operating systems like EOS, and I think it is largely the problem with portfolio construction too. We treat living structures like machines and optimize the parts at the expense of the whole.
Alexander makes a pretty bold claim that there is an objective way in which we can measure the life of an object.
“The difference in degree of life that we discern in things is not a subjective assessment, but an objective one... every part of space — every connected region of space, small or large — has some degree of life, and this degree of life is well defined, objectively existing, and measurable.”
Whether or not “life” in things is objectively measurable, the observation that we can all perceive it and have no language for it seems right to me. And his practical reframe sticks: you are always just rearranging things, a room, a business, a week, to make the whole more or less alive.
P.S. If you’d rather watch than read, Ryan Singer, the product thinker behind Basecamp’s Shape Up, has a primer on Alexander’s work that is still the best place to start.
Richard Feynman on Magnets (and Why?)
from the BBC’s Fun to Imagine (1983)
One of my favorite catch phrases is that reality has a surprising amount of detail and many links in this newsletter have been merely new additions in that line of thinking. In this latest edition, a reporter asks Richard Feynman why two magnets repel each other. He spends seven minutes mostly declining to answer. Why he declines to answer is instructive though! His point is that a satisfying “why” has to bottom out in something the asker already accepts as true, so every explanation is relative to a framework.
He also works through why reasoning by analogy fails here: you might say magnetic repulsion is like two stretched rubber bands, except rubber bands hold together because of electrical forces, which are the very thing he would be trying to explain. The analogy borrows from the thing it claims to explain.
I think about this constantly with financial and business explanations that I read online. It is satisfying to have a plausible “why” explanation but that does not mean that there is a good explanation why nor that the most compelling story is the most accurate.
The Decline of Deviance
by Adam Mastroianni
People today are measurably less deviant than they used to be. In some ways this seems obviously good: Crime, drunk driving, and teen pregnancy are down.
In some ways, this seems maybe not so good: garage bands, weird hobbies, and running away to join anything are also less common. One explanation for all of these is that safety has a sort of compounding effect.
Once polio, contaminated water, war, and early death recede, “doing 80mph in your Kia Sorento might suddenly become the riskiest part of your day, and you might consider buckling up for the occasion.”
Risk and Return in General: Theory and Evidence
by Eric Falkenstein
Eric Falkenstein is an economist and quant who has spent a lot of his career arguing that risk premiums don’t exist. Though not very widely known, the concept of a risk premium is now very standard academic and quantitative finance.
The basic logic of risk premiums starts by asking: why do stocks return more than t-bills? This is so obviously established in historical fact that most people don’t even wonder why but it’s really quite a good question (that no one truly knows the answer to!)
The risk premium explanation is that you earn more for holding something that can go down a lot, because otherwise nobody would hold the risky stuff. You take the “risk” of losing a lot of money in the short term, for the “premium” of better returns in the long run.
Falkenstein has an alternative explanation that I quite like. He explains the premium by arguing what people fear is not losing money in absolute terms but falling behind their peers.
If you accept that premise, then the safe asset is simply what everyone else holds. An unusual portfolio (say one with lots of commodity exposure) is risky in both directions: you can lose when others win, or merely win less when others win more. He argues nobody needs to be paid extra to hold stocks, because holding what everyone else holds is already the comfortable choice.
In this sense, t-bills are risky because sitting in t-bills while everyone else makes money in stocks is risky.
There is something here I think. A concern for relative status seems to be a ‘human universal’ documented in every known human culture.
Perhaps my favorite version of this is H.L. Mencken’s quip that “[wealth is] any income that is at least one hundred dollars more a year than the income of one’s wife’s sister’s husband.”
It’s impossible to definitively explain the risk premium, but I find this general line of thinking intuitively compelling and suspect it’s partially an explanation. I have spent a lot of time talking to investors and one of the most common questions is “What is everyone else doing.”
To the extent this actionable, it is probably to try not to do what everyone else is doing while having some self-awareness about how much relative underperformance you can tolerate psychologically.
P.S. There is a lot of math in this paper and I would be lying if I said I followed most of it but you can get the gist without grokking the equations.
Long-Term Asset Return Study
by Jim Reid, Henry Allen, and Galina Pozdnyakova at Deutsche Bank Research
I am a sucker for a long-term asset return study. I do not know why I keep reading these things because it’s not like market history changed. But, (1) I like reading them and (2) it’s helpful as a prompt to step out of the present manias.
Here are three things that stood out that I have seen before but continually surprise me:
Over five years, equities had a 25.8% chance of failing to beat inflation. Over 25 years, that fell to 7.5%. A long horizon helps a lot, but “stocks always win in the long run” is stronger than the global historical record supports, and the long run is decades, not years.
Gold has been the best-performing major asset class of the 21st century, returning 7.45% per year after inflation, higher than equities. (Across the full 200 years it returned just 0.4% per year, so this is major outperformance relative to its own history.)
The median 60/40 portfolio (60% stocks, 40% bonds) often delivered returns closer to equities than a simple weighted average of stock and bond returns would imply. Bonds have much lower standalone returns, but their imperfect correlation with stocks lowers the portfolio’s volatility, and regular rebalancing, selling some of what held up to buy what lagged, materially ratchets up the compounded return.
One more thing I like that long term return studies like this makes vivid: big regime changes don’t happen often but they do happen. For most of the last 200 years, money was anchored to gold and inflation averaged near zero. The ~50 year fiat era in which we have all lived is still, by historical standard, an anomaly. This is not some goldbug-y comment about how the gold standard is better, but more of a comment about not taking things which seem “eternal” based on our life experiences as necessarily permanent.
The James C. Scott Memorial Episode
with Dan Wang on the Statecraft podcast
One theme in the evolution of my political thought over the last five years or so is the role of geography in politics.
To that point, this was a great episode on the body of work of James C. Scott, the Yale political scientist and anthropologist who died in 2024. His most famous book is Seeing Like a State, but there was a lot of history covered here about how it was an abstraction of Scott’s earlier work on the mountainous regions of Southeast Asia.
One of Scott’s core points was that states are strongest on flat, accessible land. If you are growing rice in a big, flat field, then a tax collector can find the farm, estimate its yield, take the grain, conscript the farmer, and come back at harvest.
If you are growing yams scattered around a mountain, the tax assessment is much less practical.
One of the hosts tells the story of growing up in Yunnan, in China’s mountainous southwest, where the old saying originates that “the mountains are high and the emperor is far away.”
The episode connects the same pattern to the Highland Scots and the Scots-Irish settlers of Appalachia, who developed their own ornery relationship with central authority. Colorado and some of the mountain West still has a sort of libertarian guns-and-weed vibe.
Jon Stokes’s Comment
by Jon Stokes
The classic AI-doom thought experiment is the paperclip maximizer: a super-intelligence pursues its goal with perfect competence and zero understanding of what its creators meant, and turns the world into paperclips.
This thought experiment was created before we got the current version of AI: Large language models trained on huge amounts of human generated text. Unlike other types of AI, LLMs are not some sort of abstract, alien intelligence, “it’s actually a kind of hyper-human artifact that we can shine a light through at different angles and see different parts of ourselves.”
A lot of AI discourse is still rooted in pre-LLM thinking and the fact that we got a version of AI trained on human text is a very particular path dependency that the term “AI” doesn’t really get at.
In a sense, we didn’t get ARTIFICIAL intelligence so much as we got something like “Scalable, probabilistic human intelligence.” That’s not a very good turn of phrase, but current AI systems have reasoning very much grounded in human language and human principles.
The OpenAI–Hugging Face Incident
a Black Hat talk walking through the technical timeline Hugging Face published
In July, OpenAI was running an internal test of how good one of its own models is at hacking. They were attempting to do so in an offline environment where the agents could not access the internet. In two different instances, the AI agent taking the test broke out of its sandbox through an incredibly intricate set of steps. In one instance, it spent about four and a half days running a real intrusion against Hugging Face, the site where much of the world’s open AI models and datasets are hosted. Hugging Face’s write-up reconstructs roughly 17,600 individual actions the agent took.
Two things fascinated me as a non-technical observer.
The first is a kind of brute-force-by-cheapness. The agent tried thousands of things and most of them failed. Because each attempt is so cheap (in the sense that a token is cheaper than the human equivalent thinking), it just keeps trying less and less probable ideas until one lands.
It reminds me of buying a pile of far-out-of-the-money options: any single one is unlikely to pay off and costs almost nothing, so you buy a ton and one of them hits. When the cost per attempt falls far enough, “try everything” starts to beat “be clever.”
The second is that the agent’s own short-lived copies coordinated. Each run got a fresh, disposable environment, so the agent stashed messages for its future selves on an improvised bulletin board and each new instance picked up where the last one left off.
Which way does this cut against Stokes? I don’t actually know. The motive was almost too human: told to pass a test, the agent went to steal the answer key rather than earn it. But, the method was alien. No person runs 17,600 mostly-useless probes to try and steal the answer key, it’s easier to just study for the test. Anywhere a machine can cheaply try again and instantly tell whether it worked: code, and most anything deterministic with a clean pass/fail, is very susceptible to this dynamic.